Prevalence and typology of potential drug interactions occurring in primary care patients

Abstract

Purpose: To investigate the prevalence and types of potential drug interactions in primary care patients to detect risky prescriptions as an essential condition to design intervention policies leading to an improvement in patient safety. Methods: Cross-sectional descriptive study. Setting: Two areas in Spain comprising 715 661 inhabitants. Patients: 430 525 subjects with electronic medical records and assigned to a family doctor regularly updating them. Results: On a random day, 29.4% of the population was taking medication. Of these, 73.9% were at risk of suffering interactions, and these were found in 20.6% of them. The amount of interactions was higher among people with chronic conditions, the elderly, females and polymedicated patients. From the total of interactions, 55.1% belonged to the highest clinical relevance ‘A’ level, and 28.3% should have been avoided. The active ingredients primarily involved were hydrochlorothiazide and ibuprofen and, when focusing on those that should be avoided, omeprazole and acenocoumarol. The most frequent ‘A’ interaction that should be avoided was between non-conjugated excreted benzodiazepines and proton-pump inhibitors, followed by some NSAIDs and diuretics.

Conclusions: 1 in 20 Spanish citizens is currently undergoing a potential drug interaction, including a high rate of clinically relevant ones that should be avoided. These results confirm the existence of a serious safety issue that should be approached and where all parties involved (physicians, health services, medical societies and patients) must do our bit to improve. Health services should foster the implementation of prescription alert systems linked with electronic medical records including clinical data.

Key words::

• Drug safety
• drug interactions
• community practice
• computer technology

Introduction

Drug prescription constitutes a complex daily decision-making process for physicians and is subject to great variability. In fact, the prescription of drugs is the most common outcome of patient consultations and, thus, one of the main causes of the occurrence of adverse events in primary care settings (1–3).

The incidence of medical errors and their consequences have traditionally been under-estimated, since many of them go unnoticed, do not produce directly recognizable harm, or are solved in a timely manner (4). The report issued by the Institute of Medicine ‘To Err is Human’ and subsequent studies have highlighted the magnitude and importance of primary care, frequently forgotten in favour of the hospital setting (3,5–7). In general, errors due to the use of medications are the most relevant after diagnose-related errors and mainly occur in chronic diseases or in elderly patients (8–12). The APEAS study estimated a prevalence for adverse events of 11.18 per 1000 consultations in the Spanish primary care setting, with the most frequent of these being medication-related (5.4 per 1000 consultations) (3). In the United States, the incidence of medical errors ranges between 0.05 and 0.8 per 1000 consultations, with medication-related errors as the second cause. These are commonly due to dosing issues or to the occurrence of interactions, which may be defined as a change in the response to a drug as a consequence of a previous or concurrent administration of another drug (1,13).

Although drug interaction analysis is often routinely included in quality assurance programmes, most interactions go unnoticed by physicians due both to the absence of new clinical signs and symptoms and because they often produce a worsening of already treated diseases (14,15). Furthermore, if new signs and symptoms appear, they are often confused with a new disease that leads to prescription of additional drugs, thus increasing the likelihood of the onset of interactions (8).

In developed countries, increased life-expectancy in combination with both a high prevalence of chronic conditions and the high number of patients simultaneously taking several drugs, have been described as important predictors for interaction occurrence (12,16–19). Furthermore, the frequent presence of multiple physicians involved in patient management (in addition to family doctors) increases the risk of inappropriate drug combinations (20–22). In this way, large studies conducted in the European Union indicated a prevalence of interactions ranging from 12% to 15% (23–25). The Spanish situation is not well known, although the limited studies available indicate prevalence rates between 13.6 and 31.5% (8,26).

The goal of our study is to investigate the prevalence and types of potential drug interactions occurring in primary-care treated patients in Spain. It is not aimed at describing the interactions actually affecting them, but only to detect risky drug prescriptions as an essential prior condition for the design of active intervention policies leading to an improvement in patient safety.

Methods

Design

We designed a cross-sectional, descriptive study in health areas I and VI of the Region of Murcia, in South-eastern Spain. In this region, OMI-AP is the electronic medical records system currently applied. It manages all relevant clinical data: notes on patient’s consultations and their conditions, list of problems, clinical orders and courses of treatment, including personal plans and approved clinical practice guidelines. It is also provided with modules for managing sick leave and both acute and chronic prescriptions. OMI-AP data are available from 715 661 people out of a total population of 723 664 inhabitants (14.7% aged more than 65, and 50.4% was female). Among these, people older than 14 are assigned to a family doctor.

430 525 subjects from this population were included in this study. They met the following criteria: They had OMI-AP data available, were older than 14 and they were assigned to a family doctor regularly updating records (this definition includes all doctors producing over 100 electronic prescriptions per week as indicative of an accurate use of OMI-AP electronic medical records). Of these, 15.0% were aged more than 65 (95% confidence interval, CI: 95 ± 0.1), whereas the rate in the total population is 14.7%; and 51.0% was female (CI: 95 ± 0.2; total population rate 50.4%).

As the study was designed to deal with electronic records that can be processed automatically, no sample was extracted from the population meeting the criteria set out above. Thus, all medications registered in OMI-AP that all these patients could be taking on a randomly selected date, 7 March 2007, were analysed. As in Spain family physicians are currently signing most of the prescriptions given in our health service (even those issued by other physicians, e.g. working in a hospital), in practice this represents their entire pharmacological treatment.

Measurements

We designed and tested software in collaboration with the Information Technology Department of Murcia Health Service. This software, especially built for this purpose, is able to review all concurrent prescriptions from each patient at a definite time-point using both the OMI-AP electronic medical records and the specified treatment duration (if unavailable, it is estimated from the number of defined daily doses prescribed and the prescription date). Medications were deemed concurrent if the duration of therapy for one drug (calculated according to dosage instructions and quantities prescribed) overlapped the duration of therapy for the other drug. All active substances that any given patient might be taking simultaneously were analysed for interactions.

Interactions were collected and classified in accordance with the medications database (BOT) of the General Council of Official Colleges of Pharmacists in Spain (27). A double criterion was followed:

1. Relevance of the interaction: We distinguished four levels in decreasing order, from (A) to (D), based on the nine original groups in BOT:

   (A) Relevant interaction (BOT groups 1–2): Clinically significant interaction that always appears when the medications are administered concurrently.

   (B) Relevant interaction under special circumstances (BOT groups 3–4): evident in certain patients due to dose-dependence, pathologic status, etc.

   (C) Potentially relevant interactions, but with no adverse events (BOT groups 5–6): Proven interactions, although no adverse events have been described.

   (D) Low relevance or theoretical interaction (BOT groups 7–9): Limited clinical significance, or only theoretically described.

2. Remedial action: Interactions were also assessed in order to know what should be done after their identification. We distinguished three different actions:

   0. Interactions to be avoided: action must be taken to avoid the appearance of adverse events.

   1. Interactions requiring surveillance: require surveillance or monitoring of patients.

   2. Interactions requiring a modification of the dosing interval, usually an increase of the time period between the doses of interacting drugs.

By combining these two criteria and paying attention to the specific active ingredients involved in each case, a final operational classification was built. This was essential because some interactions are characteristic of a specific therapeutic group or subgroup, while others only involve or have different severity levels when specific active substances or a specific substance interacts with a specific therapeutic group or subgroup. This operational classification attempts to reflect all these peculiarities, thus preserving the relevance of the interaction and the remedial action with the greatest possible accuracy. For example, the interaction between non-steroidal anti-inflammatory drugs (NSAIDs) and diuretics has been operationally classified in different ways (A0, A1, D1), depending on the type of NSAID and diuretic involved.

Statistical analysis

The different types of potential interactions found by the software and several patient-related factors were investigated for patients that could be simultaneously taking more than one drug and therefore at risk of suffering interactions. These factors included age, gender, medications taken, and presence in OMI-AP data of a selection of chronic conditions from the Murcia Health Centres’ Services Chart (hypertension, dyslipidemia, diabetes, chronic obstructive pulmonary disease, asthma and degenerative osteoarthritis) (28). The prevalence rates and their corresponding confidence intervals were calculated, using odds ratios obtained with a forward stepwise method of logistic regression with entry-critical P value equal to 0.05 and removal P value equal to 0.10, and the appropriate formulas were applied (29).

Results

Study population and prevalence rates

The study found 451 175 active prescriptions in 126 414 patients, indicating that 29.4% (CI: 95 ± 0.1) of the population was taking some kind of medication on the selected study day. Of these patients, 93 481 (73.9%, CI: 95 ± 0.2) were simultaneously taking more than one drug (median 4; mean 4.7) and were therefore at risk of suffering interactions.

33 638 interactions were detected in 19 279 patients—20.6% (CI: 95 ± 0.2) of at-risk patients, and 15.2% (CI: 95 ± 0.3) of medicated patients—i.e. a mean of 0.36 interactions/patient at risk (CI: 95 ± 0.01). A higher probability of an interaction appearing was found among people with chronic conditions, the elderly, females, and polymedicated patients (Table I).

Table I. Presence of potential interactions according to different patient-dependent factors. Ratio of adjusted prevalence and confidence interval.

	Presence of interactions (%)
	Exposed	Non-exposed	Prevalence ratio	95% Confidence interval	Significance (P)
Number of medications prescribed^a	39.7	7.5	4.70	4.56–4.86	<0.0001
Presence of a chronic disease	26.8	11.8	1.54	1.49–1.60	<0.0001
Advanced age (over 65 years)	28.5	14.4	1.19	1.15–1.23	<0.0001
Gender, females	21.3	19.3	1.07	1.04–1.10	<0.0001

^aNumber of medications prescribed: exposed and non-exposed groups are differentiated according to their median (four prescriptions).

Interaction analysis

By relevance, the most important ‘A’ type interactions were the most frequently observed (18 542 cases, 55.1% of total, CI: 95 ± 0.5), whereas, according to the recommended course of action, 9523 interactions (28.3%, CI: 95 ± 0.5) should have been avoided. Of these, 6227 were also ‘A’ level, representing 18.5% of the total (CI: 95 ± 0.4). A complete distribution of the interaction types identified is shown in Table II.

Table II. Number and type of potential interactions identified.

			Remedial action
			0Avoid association	1Patient surveillance	2Increase time interval	Total
Relevance	A	Relevant	6 227	9 949	2 366	18 542	55.1%
	B	Relevant in special circumstances	1 020	2 143	0	3 163	9.4%
	C	Potentially relevant	1 298	3 545	39	4 882	14.5%
	D	Low relevance	978	6 073	0	7 051	21.0%
		Total	9 523	21 710	2 405	33 638
			28.3%	64.5%	7.1%

Again focusing on ‘A’ level interactions that should be avoided (A0), the active substances most frequently involved were omeprazole (4.3% of prescriptions and 1 164 interactions found, 9.3%), acenocoumarol (0.6% of prescriptions and 838 interactions found, 6.7%) and diazepam (0.7% of prescriptions and 822 interactions found, 6.6%). In the interactions identified as a whole, hydrochlorothiazide (2.4% of prescriptions and 5 499 interactions found, 8.2%) and ibuprofen (2.4% of prescriptions and 3741 interactions found, 5.6%) were the ones most frequently involved.

The most frequent type A0 potential interaction was observed between non-conjugated excreted benzodiazepines and proton-pump inhibitors (3.4%), where there was a risk of benzodiazepine toxicity, followed by the interaction of certain non-steroidal anti-inflammatory drugs (NSAIDs) and diuretics (2.1%) with risk of diuretic effect loss. As a whole, the most frequent operational interactions were this last one (10.0%) and the type A2 interaction between calcium salts and bisphosphonates (6.0%). Operational interactions identified with a frequency over 1% are shown in Table III.

Table III. Potential operational interactions with a frequency >1%.

Interaction: Operational type	%	% Ac.	Comments
NSAIDs + diuretics—D1	10.0	10.0	Decrease of the diurectic/antihypertensive effect due to inhibition of renal cyclooxygenase. The NSAIDs involved have limited affinity for this enzyme (for example, selective inhibitors of COX2), and/or are combined with a diuretic producing an interaction with no clinical relevance evidence.
Calcium + bisphosphonates—A2	6.0	15.9	Their concurrent administration may reduce bisphosphonate serum levels due to a decrease of it's oral absorption. These medications should be taken with at least one-hour margin between them.
NSAIDs + ACEI—D1	5.3	21.3	Decrease of the antihypertensive effect of ACEI by inhibition of prostaglandin synthesis at renal level. Excludes associations of NSAIDs and ACEIs with significant and well-documented interaction, for example with indomethacin or ibuprofen with captopril.
Inhaled selective beta-2 agonists + diuretics—A1adrenergic	4.7	26.0	Reinforcement of the hypokalaemic action of beta-2 drugs by non-potassium sparing diuretics, with risk of tachycardia and other dysrhythmias.
Clopidogrel + ASA (cardio)—C1	3.8	29.8	Possible reinforcement of the collagen-induced platelet-aggregation inhibitory effect of ASA. Although no clinical evidence is available, the safety of the concurrent administration of these medications has not been established, and caution is recommended when used concomitantly.
Benzodiazepines + Omeprazole—A0^a	3.4	33.2	Omeprazole inhibits the hepatic metabolism of benzodiazepines at the P450 cytochrome level. When used in association with benzodiazepines requiring this metabolic step(diazepam, bipotassic chlorazepate, etc.), it leads to an increase in serum levels and possible intoxication.
Diuretics + digoxin—A1	2.8	36.0	Possible increase in digoxin toxicity when combined with non-potassium sparing diuretics, leading to a decrease in potassium serum levels.
Calcium + thiazides and analogues—A0^a	2.3	38.3	Possible accumulation of calcium in the body due to inhibition of its urinary excretion that may lead to toxic effects.
Acenocoumarol + Omeprazole—A1	2.3	40.6	Possible reinforcement of the anticoagulant effect due to inhibition of hepatic metabolism.
NSAIDs + diuretics—A0^a	2.1	42.7	Loss of the diuretic/antihypertensive effect due to renal cyclooxygenase inhibition. There is extensive clinical evidence of interaction with the NSAIDs involved (for example, indomethacin) and/or combined with potent diuretics (for example, loop diuretics).
Inhaled selective beta-2 agonists + glucocorticoids—A1	2.1	44.8	Reinforcement of the hypokalaemic action of beta-2 adrenergic drugs by glucocorticoids, with risk of tachycardia and other dysrhythmias.
Sulfonylurea + thiazide and analogues—B1	2.1	46.9	Decrease of the hypoglycaemic effect of sulfonylureas, and risk of hyperglycemia due to a diminished insulin release as a consequence of the hypokalaemic effect of thiazides.
Allopurinol + thiazides and analogues—A1	1.8	48.7	Occasional occurrence of significant allergic reactions of unknown mechanism.
Angiotensin II antagonists + potassium sparing diuretics—D1	1.3	50.1	Possible occurrence of toxicity due to hyperkalaemia, by addition of the potassium-sparing effect of both drug classes.
Digoxin + proton pump inhibitors—C1	1.3	51.4	Possible accumulation in the body and toxicity of digoxin that may be due to increased gastrointestinal absorption as a consequence of the increased gastric pH produced by proton pump inhibitors.
Acenocoumarol + ASA (cardio)—A0^a	1.2	52.5	Reinforced anticoagulant effect with risk of bleeding. The association must be avoided, except in selected patients for whom a consensus has been reached, for example those with cardiac prosthesis and suffering CVA in spite of an adequate INR.
Benzodiazepines + serotonin selective reuptake inhibitors—A1	1.1	53.7	Possible increase of plasma levels and benzodiazepine toxicity due to inhibition of the hepatic metabolism when associated with certain serotonin selective reuptake inhibitors, included fluoxetine or fluvoxamine.

^aA0 type interactions.

% Ac: accumulated percentage.

NSAIDs, non-steroidal anti-inflammatory drugs; ACEIs, angiotensin converter enzyme inhibitors; ASA (cardio), anti-aggregative doses of acetylsalicylic acid.

Discussion

We found a drug interaction in about 20% of primary care patients taking more than one drug on a random date, i.e. one out of every twenty people in Spain are subjected to some kind of drug interaction right now. Furthermore, 55% of these come from the most relevant A-type and 28% should have been avoided. This problem is even more important among people with chronic conditions and the elderly, thus confirming the existence of a serious patient safety issue that should be dealt with as soon as possible.

Strengths and limitations of the study

The available studies tackling the problem of potential drug interactions in patient safety from an epidemiologic/population perspective are limited. This study analysed the prescription patterns in a population using existing data in OMI-AP software. OMI-AP is a practically universal system within our Health Service, widely used by family doctors as indicated by the great rate of the overall population who met inclusion criteria (60.2%) and their high similarity in both age and gender rates (albeit there are some statistical differences between the sample and the total population, these are not clinically significant and probably attributable to its large size). This contrasts with other studies based on electronic medical records that have a lower sample size or are based on prescriptions dispensed in community pharmacies (9,12,23–25,30). In Spain, community pharmacy data are, in general, not linked with the patients and their clinical data because they only identify the drug dispensed and the physician prescribing it. Using OMI-AP data enabled us to investigate the relationship between prescription and some clinical features (as chronic conditions, etc.) from an entire population perspective, and paved the way for future research and improvement.

As information about prescription systematically offered to Spanish physicians is usually generic, not linked to the clinical features of their patients, and does not include any alert system, another added value of the study could be the design of a software able to cross all concurrent prescriptions from a particular patient with a database containing the interactions classified according their clinical relevance and the recommended course of action. This information could be very useful to our family doctors because they could be alerted if an important potential interaction were detected. In this way, the real patient status could be checked and, if necessary, the suggested remedial actions taken. We think this could improve their medicated patients’ safety.

The nature of the database for identification of interactions, as well as the way interactions are categorized, may affect the comparability of studies like this. Unfortunately, this is highly variable and occasionally investigators fail to disclose the method used (24,31). This research is based on the classification established by the General Council of Official Colleges of Pharmacists in Spain for its medications database (BOT) (27). It is similar to the well-known Hansten & Horn classification and its commonly used variations (19,23,25,32). This database is widely used in our country, its Internet version is updated daily and has been gradually implemented within the health environment with good structural quality criteria, even better than Hansten is (33,34). Drug interactions are widely documented with updated references. This database also provides information on the attitude to be adopted by the clinician in front of a particular interaction, offering, when necessary, a safer option. Since there may be a lack of consistency in the inclusion and grading of drug interactions between different classifications, our group compared its parameters with other databases of recognized quality, as the Thomson Micromedex and others to design a definitive classification (35–40).

The classification eventually adopted in this study summarizes the nine original BOT relevance groups in only four categories (A to D) because we have merged those groups that having a similar clinical relevance only differ in the volume of their evidence references. We have also added the recommended action and secondary operational classification in order to provide more useful information to our family doctors.

Magnitude of the problem

As mentioned above, the various criteria used to categorize interactions result in a great variability of drug interactions published results. This way, although the rate of interactions per total number of prescriptions was within the range found by other authors (2–14%) (23,24,31), the present study indicated a higher percentage of subjects was taking over one medication at the time (74% versus 19–64%) and, as a result, a higher rate of interactions (20.6% versus 12–15%) (9–11,23–25,31,41). As an exception, the Rotterdam Study observed a similar prevalence although this study was conducted in people over 55 years of age (42).

Risk factors

As did most authors, this study also observed a direct association between advanced age and the amount of concurrent medications with the risk of suffering a drug interaction (9,19,23,24,43) (Table I). On the contrary to the Linnarson study, we identified a higher risk among females (23). This might be partially due to the wide size of the sample used, able to find few risk differences that were statistically significant.

We also verified an increased risk in the presence of a chronic condition. In general, most studies had not analysed this factor probably because of the non-availability of this kind of data. The access in this study to electronic medical records containing clinical information has made it possible to analyse this connection, which must be confirmed in further studies.

Knowledge of these risk factors may make it easier to avoid their appearance as confounders in further studies aimed to check the effectiveness of the improvement programmes that should be put into place following the results of this study.

Interaction analysis

Regardless of the attitude the physician should adopt, 55.1% of the potential interactions identified were considered clinically relevant (‘A’ level), a value much higher than those reported by other authors. This difference might be due both to the methods used to identify interactions and to a possible overestimation of interactions in the present study.

In this sense, the software used for the study considers all active prescriptions found in OMI-AP to be effective. Nevertheless, some of these could have been included in its automated system for chronic patients’ repeat prescriptions and mistakenly considered as active when in fact they had been discontinued. There again, the present study only knows if a prescription was given by a doctor, but not whether the pharmacy filled the prescription, or if the patient stopped taking a drug that could interact with it, etc.

We consider the influence of these possible biases would be minor because of the high correlation observed between the OMI-AP active prescriptions and those filled in pharmacies. Moreover, the focus of this study on potential interactions and patient safety takes this approach to the problem into account. As said before, this study deals with detecting potential interactions within doctors’ intention to treat, rather than real or clinical threats.

With regard to the active substances and pharmacologic groups involved, a wide correlation with other studies was observed, showing a high percentage of interactions among diuretic drugs, NSAIDs, anticoagulant drugs, systemic beta-blockers, digoxin, angiotensin converter enzyme inhibitors, etc. (12,23–25) (Table III). Most of these drugs are used to manage chronic conditions worldwide, in particular cardiovascular risks.

Furthermore, the present study found relevant ‘A’-level potential interactions that should be avoided and are not mentioned in other studies, including interactions between certain benzodiazepines and omeprazole, calcium and thiazides, or NSAIDs and certain diuretic drugs (Table III). On the one hand, these differences might be due to a greater prescription of proton pump inhibitors in our country. On the other hand, the cases of thiazides plus calcium and NSAIDs plus diuretics could be explained because other studies might have not considered an operational classification and thus not have categorized some of the interactions shown in Table III as clinically relevant (‘A’ level) as the present study does.

Implications for practice and conclusions

As previously stated, our study suggests that 1 in 20 Spanish citizens is currently subjected to a potential drug interaction, including a large number of clinically relevant ones that a priori should be avoided. These results confirm the existence of a serious patient safety issue that should be dealt with as soon as possible and where all parties involved (physicians, health services, medical societies and patients) must do our bit to improve.

In our opinion, health services should foster the implementation of alert systems on prescriptions linked with electronic medical records including clinical data. These could make it easier to plan and implement improvement initiatives aimed at detecting and preventing prescription problems in the primary-care setting.

Declaration of interest: The authors report no conflict of interest. The authors alone are responsible for the content and writing of the paper.